The Agent Skills Directory

[COMMAND_EXECUTION]: The script utilizes the source command to dynamically load profile configuration files from ~/.config/sqlite-crud/profiles/. While the skill validates profile names to prevent path traversal and sets restrictive file permissions on the config directory, loading shell scripts from computed paths is a sensitive pattern that could lead to code execution if the configuration files are tampered with.
[COMMAND_EXECUTION]: The skill constructs SQL queries by interpolating user-provided identifiers and parameters. Although it includes logic to quote identifiers using a regex-based whitelist and escapes literals by doubling single quotes, the dynamic assembly of SQL commands based on external input remains a significant security surface for potential injection or logic bypass.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes data from local SQLite files (via DB_PATH). While it uses identifier validation and literal escaping to sanitize SQL construction, there are no boundary markers to delimit or ignore instructions potentially embedded within the database content itself. The skill's capabilities include executing subprocesses through the sqlite3 CLI and writing results to the local filesystem, which could be leveraged if the agent is manipulated by data residing in the processed databases.

sqlite-crud